Research On Speech Enhancement Based On Dual Microphone Array

In modern communication,multimedia technology,human-computer interaction,and smart home system,due to the background noise and reverberation interference,microphone receives the signal is usually noisy.This is not only affects the speech intelligibility,but also affects the overall performance of the speech processing system.Therefore,it is necessary to reduce the noise.Compared with single microphone array,microphones can make full use of the space-time information of speech signals and make a good performance in noise reduction.In addition,with the development of technology,electronic equipment is increasingly tend to be miniaturization.As a microphone array speech acquisition device,its volume cannot too big,and not too much.Based on the above background,this paper mainly studies dual microphone array speech enhancement algorithm,the main work includes:(1)Summarized near and far field model of the microphone array,and its topology structure.And on this basis,this paper introduces the speech signal preprocessing methods and the characteristics of noise,and gives the evaluation of speech quality and intelligibility of subjective and objective evaluation method.(2)Put forward a improved noise cross-power spectrum estimation algorithm.In conventional noise estimation algorithms,VAD algorithm does not applied to the non-stationary noise environment,MT algorithm cannot fast response and update the noise power spectrum estimation and IMT algorithm noise estimation inaccurate.Proposed method using recursive average technology that smooth factor will be based on the actual signal-to-noise ratio of each frequency point to change adaptive;Simulations show that the proposed noise power spectrum estimate algorithm is superior to each IMT algorithm,and computational complexity does not increase.(3)Put forward a harmonic phase estimation method which based on the fundamental frequency and signal-to-noise ratio.In dual microphone array speech enhancement algorithm,typically the noisy spectral amplitude is modified while the noisy phase is used to reconstruct the enhanced signal.Considering the phase difference between the voice and pure voice,we decompose the instantaneous phase phase into two parts: unwrapped phase and linear phase.The smoothed unwrapped phase combined with linear phase to form enhanced phase.Simulations show that the proposed method can effectively improve the performance of the system.(4)Put forward a joint phase and amplitude spectrum estimation method which combined improved noise cross-power spectrum estimation algorithm with harmonic phase estimation method.On one side,the combined method can get the amplitude spectrum as well as the phase spectrum of speech.On the other side,it can modified amplitude spectrum and phase spectrum of noise in time.When compared with the improved noise cross-power spectrum estimation algorithm and harmonic phase estimation method,the joint method can obtain higher speech quality.